Research · the cost of doing it yourself

Self-hosting is the expensive option.

The engine underneath big-file version control is open. You can run it yourself, and the arithmetic looks obvious before you do it: object storage is a couple of cents a gigabyte, so surely running your own is cheaper than paying anyone. It comes out higher, and the gap is not storage.

Where the estimate goes wrong

Everyone prices the bucket. Nobody prices the other three lines.

It is not a bucket. The engine does not keep a folder of files somewhere and call it a day. It keeps an index: what fragments exist, which version points at which of them, what has been superseded. That index lives in its own database, and it is not optional — the service will not start without one. So the shopping list starts at two systems, not one, and the second is the sort of thing that is cheap to run and expensive to get wrong.

The bytes go through your server. This is the line that decides the bill. In the open design, a download is read by the server and handed to the client, which means your machine is on the path of every gigabyte anyone pulls. You do not size that box for how many people are using it. You size it for the bandwidth it has to move, and it has to be awake at three in the morning in case somebody syncs.

Then you pay for those bytes twice. Once in the server you sized to move them, and again in egress, which is the line item every cloud is quietest about. It is roughly nine cents a gigabyte once past a small free allowance. A single artist pulling a single one-terabyte project, once, is most of a hundred dollars on its own.

The monthly bill

One terabyte. One full pull. Every month.

A single project, a terabyte on disk, downloaded in full once during the month. That is a small studio with one editor and one render machine, not a busy one.

Rounded, and rounded in self-hosting's favour wherever a range existed. Cloud list prices move and hardware gets cheaper; an argument that needs the worst case to win is not an argument. Ours is the published price for a terabyte, with no seats, no per-user line and no minimum.

And then the real problem

You have bought a storage engine, not a product.

This is the part the cost table cannot show. Everything above buys the layer that keeps bytes and versions straight. That layer is genuinely good, and it is genuinely all it is. A version control engine has no opinion about people.

Every one of those is a thing you would build, and then maintain, and then be on call for. None of them is interesting work. All of them are the difference between a storage engine and something a studio can actually use on a Monday.

The line item with no price

The fork is the bill that never stops.

To make an open engine into something usable you will patch it, because the gaps above are not configuration. Patch it and you own a fork. Own a fork and every upstream release is a merge, and every merge is a weekend, and the weekends do not appear in any table.

The honest version of the self-hosting question is not can I run this. You can. It is do I want a second product to maintain, in order to pay more for the half of it that already exists.

What we actually sell

One line on the bill, and the product is already built.

Billed in blocks, no seats and no per-user tax at any team size — the owner of the space pays and everyone else just works. Unlimited projects. The desktop app, whole-machine search and the ifile CLI are free forever on every machine you own, and always will be; hosting is the only thing with a price on it.

The comparison is not really a price comparison. It is one bill against another bill plus a system to run, for a smaller result. That is why the cheap-looking option is the expensive one.